关于Slurm中pam_slurm_adopt的action_unknown里source job含义的咨询

Great question! Let's break down what this term means in the context of Slurm's pam_slurm_adopt module:

Core Definition of "Source Job" #

The source job refers to the specific job that the pam_slurm_adopt module expects to tie your login session (typically an SSH connection to a compute node) to. When you log into a node where you have active jobs running, the module uses RPC calls to the Slurm controller to identify this intended job—this is the source job. It's the job you'd logically want your session associated with, so your session inherits that job's resource limits and cgroup constraints.

When action_unknown Triggers #

This configuration comes into play when two key conditions are met:

• You have multiple active jobs running on the target node
• The RPC request to the Slurm controller fails to locate that intended source job (this could happen if the controller is temporarily unreachable, or there's a mismatch that prevents it from pinpointing the correct job)

What Each action_unknown Option Does #

Per the official documentation, you have three choices for handling this scenario:

• newest (default): Selects the most recent job on the node, determined by the modification time (mtime) of the job's step_extern cgroup. This requires memory cgroups to be enabled. Your SSH session will be attached to this job, though it might not be the one you intended (e.g., if the job you wanted already exited).
• allow: Permits your connection to go through without attaching it to any job—meaning no cgroup or resource limits are applied to your login session.
• den: Blocks your connection entirely.

内容的提问来源于stack exchange，提问作者saladi